Models_from_Scratch

Goal

In this problem we consider building classifiers for a 15-way fruit classification task. The dataset we consider is comprised of a set of 256 × 256 × 3 (256 × 256 pixels and 3 color channels) images of various types of fruit. The objective is to use the provided data to come up with a classifier that achieves high accuracy on a test set.

Dataset

You can get the datasets from this google drive: Link. It is one zipped file called datasets.zip containing both the train and validation data, and that is about 1.8GB big. After this data is uncompressed, it creates a directory called ./datasets that contains the .pkl files that, together, are about 8.6GB big. The training set contains n = 9, 876 image samples while the validation set contains n = 2, 116 samples. In each dataset, we provide the following:

• images: the raw RGB images
• feats: m = 15 dimensional real-valued condensed representations of each image (these were precomputed by using a pretrained neural network for feature extraction followed by PCA to reduce the dimensionality down to m = 15)
• labels: an integer value between 0 and 14 giving the class label. The integer to class mapping is provided below in the variable idx to class. The test set will include both a set of images and a 15-dimensional featurization of those images in exactly the same form as the features that were given for the train and validation set. It is available in the same link.

Scratch Model

Models like Logistc Regression, Decision Trees, Random Forests and K-Nearest Neighbors are implemented from scratch in the notebook. Along with the scratch models, bunch of other models were also implemented.

Usage

For a quick glance, click on the fruit_classification.ipynb which has all the code and outputs with explanations. To reproduce follow the below usage steps:

1. Open your terminal
2. Clone the repoistory using git clone [email protected]:nvmcr/Models_from_Scratch.git.
3. This will create a repository with the name Model_from_Scratch.
4. Enter the present repo using cd Model_from_Scratch/.
5. Set up a new virtual environment with all necessary packages and dependencies using conda env create -f environment.yml
6. Activate the virtual environment with conda activate ee-511.
7. Open the notebook, explore and run.
8. You can deactivate the virtual environment later using conda deactivate